Efficient lossless memory compression method and system

Abstract

The invention discloses a high-efficiency lossless memory compression method and system. The method comprises the following steps: acquiring an original compression signal input by the system; the original compressed signals are directly stored in a memory, and a large number of caches are established on the time dimension; when a subsequent unit needs data, the compressed signal stored in the memory is subjected to instant use and instant solution processing, and the data needed after decompression is stored in a temporary space VBUF; and the signal processing unit reads the required signal from the temporary space VBUF for corresponding signal processing. According to the invention, the input compressed data is directly multiplexed, and on the premise of ensuring that the data is lossless, the ultra-high compression efficiency is provided through the design of using and solving, so that the memory storage space and bandwidth are effectively reduced.

Description

Technical field [0001] The present invention relates to the field of memory compression technology, in particular to an efficient and lossless memory compression method and a system thereof. Background [0002] as Figure 1 As shown in traditional video and image processing systems, the bandwidth and storage capacity of memory are more than enough relative to processing tasks, so the intermediate results of the general system are stored directly into memory, as shown on the left side of the following figure. [0003] as Figure 2 As shown, but in systems with high data throughput, such as graphics card GPUs, video image AI systems, the image data cached inside the system tends to occupy a large amount of memory space, making memory bandwidth and memory capacity called bottlenecks. It is worth noting that this video and image data often derives itself from video and image content in various compressed formats. Therefore, in order to save memory storage space or reduce memory read ba...

AI Technical Summary

Problems solved by technology

This brings three problems: 1) When data is written, an additional data compression unit needs to be designed to increase the write delay; 2) Considering that data needs lossless compression, the amount of compressed data is much larger than the original input compressed data

Even for the first type of compression technology designed specifically for image data, the lossless compression can only achieve an average of 2 times, while the original input compressed data can achieve hundreds of times of compression; 3) The compression algorithm cannot be compared with the original system's inherent image or Multiplexing of video decompression units increases the cost of chip design in terms of IP authorization and area

Images